Vehicle Recognition Algorithm in Road Scene Based on YOLOv8

In the field of vehicle driving, the popularization of intelligent driving technology is the trend of the times. Many program models are applied to solve the problem of external environment perception in intelligent driving technology, which has different programs for different sensors. In this paper, YOLOv8 model is used to identify vehicles in real-time in road scenes. YOLOv8 is a convolutional neural network model composed of Conv, C2F, and other convolution operations used for target recognition. The multi-layer output is used to operate multi-layer convolution to help the model effectively identify images of different levels and reduce the invalid recognition and false recognition caused by the recognition target not being at the same model level. For example, in YOLOvX, decoupled output heads perform regression learning for categories and bounding boxes. The detection, segmentation, and pose models of YOLOv8 are pre-trained on the COCO dataset, while the classification model is pre-trained on the ImageNet dataset. The models are downloaded from the latest Ultralytics release upon first use. YOLOv8 offers 5 model sizes, allowing developers to balance performance and accuracy. Through the demonstration of the structure, program application, and effect of the model, the superiority and necessity of the application of the model can be demonstrated in detail.